Recommended system needs based on historical user behavior and interests of users predict future behavior and interests, and therefore a large number of user behavior data has become an important part of the recommendation system and prerequisites. For many, like Baidu, Dangdang such sites, or this
promise is not a problem, because they already accumulated a large amount of user data. But for many sites do purely recommendation system (such as Jinni and Pandora), or a lot of hope at the beginning of the personalized recommendation applications have web sites,
How not design a large number of user data under circumstances personalized recommendation system and allow users to recommend satisfactory results so willing to use the recommendation system, the problem is the cold start.